Non-volatile memory arrays, such as erasable, programmable read only memory (EPROM) or flash memory arrays, or electrically erasable, programmable read only memory (EEPROM) arrays, require high positive or negative voltages to program and erase memory cells of the array. Typically, these voltages are higher than the voltage supplied (Vdd). Charge pumps are generally used to boost on-chip voltages above the supply voltage Vdd to reach the voltages required for programming or erasing.
A charge pump typically comprises cascaded stages that progressively boost the voltage to higher levels. The charge pump functions by progressively storing more charge on a capacitor which is part of a capacitor-diode combination, with several such stages being placed together in a network to obtain the desired increase in voltage. The diode functions to prevent discharge of the capacitor prior to placing the additional charge thereon.
Reference is now made to FIGS. 1A and 1B, which illustrate a commonly used charge pump architecture, called a four-phased-clock, threshold-voltage-canceling pump architecture, for a four-stage charge pump (see Umezawa, IEEE Journal of Solid State Circuits, Vol. 27, 1992, page 1540). FIG. 1A illustrates two stages of the charge pump in greater detail than FIG. 1B, which illustrates four stages of the charge pump.
The charge pump circuit includes a plurality of charge transfer transistors (reference letters mi) connected in series. In FIG. 1B, four such charge transfer transistors are shown, labeled m1, m2, m3 and m4. Charge transfer transistors mi may use, but are not limited to, CMOS (complementary metal oxide semiconductor) technology, being either n-channel or p-channel (NMOS or PMOS) field effect transistors (FETs). (As explained further hereinbelow, NMOS is generally used to pump positive voltages, whereas PMOS is generally used to pump negative voltages.) The MOSFETs have a control electrode (gate, labeled g), a first electrode (drain, labeled d) and a second electrode (source, labeled s), connected to nodes, as described hereinbelow. (Since MOSFETs are typically symmetrical components, the true designation of “source” and “drain” is only possible once a voltage is impressed on the terminals of the transistors. The designations of source and drain throughout the specification should be interpreted, therefore, in the broadest sense.) Preferably, the bulks (labeled b) of the charge transfer transistors mi are coupled to a reference line (shown as REF in FIG. 1A, but omitted for the sake of simplicity in FIG. 1B) for receiving a reference voltage, generally ground in the case of NMOS.
FIGS. 1A and 1B illustrate a positive charge pump based on NMOS. The source of charge transfer transistor m1 is connected to node n0, which is connected to Vdd. The gate of charge transfer transistor m1 is connected to node g1, and the drain is connected to node n1. The source of charge transfer transistor m2 is connected to node n1, the gate is connected to node g2, and the drain is connected to node n2. Similarly, as shown in FIG. 1B, the source of charge transfer transistor m3 is connected to node n2, the gate to node g3, and the drain to node n3. Likewise, the source of charge transfer transistor m4 is connected to node n3, the gate to node g4, and the drain to node n4.
Two-phase, non-overlapping pulse trains PH1 and PH2 are provided, such as from a pulse generator (not shown). By non-overlapping it is meant that 0 to 1, and 1 to 0 voltage transitions of one pulse never overlap with the transitions of the other pulse. The PH1 and PH2 phases inject energy into the pump through large capacitors 5 into nodes ni. Accordingly, in the illustrated embodiment, a large capacitor 5 is connected from pulse train PH1 to node n1, and another large capacitor 5 is connected from pulse train PH1 to node n3. Another large capacitor 5 is connected from pulse train PH2 to node n2, and another large capacitor 5 is connected from pulse train PH2 to node n4. The charge is transferred along the pump through charge transfer transistors mi connecting node ni to node ni+1.
Similarly, two-phase, non-overlapping pulse trains PH1A and PH2A are also provided. The PH1A and PH2A phases inject energy into the pump through small capacitors 11 into nodes gi. Capacitors 11 preferably have a much smaller capacitance than large capacitors 5. In the illustrated embodiment, a small capacitor 11 is connected from pulse train PH1A to node g2, and another small capacitor 11 is connected from pulse train PH1A to node g4. Another small capacitor 11 is connected from pulse train PH2A to node g1, and another small capacitor 11 is connected from pulse train PH2A to node g3.
As seen in FIGS. 1A and 1B, a plurality of auxiliary transistors ti (i.e., t1, t2, t3 and t4) are provided. Each auxiliary transistor ti has its drain connected to the gate node gi of each charge transfer transistor mi (i.e., m1, m2, m3 and m4, respectively). The source of each auxiliary transistor ti is connected to the source of each charge transfer transistor mi (i.e., node ni−1). The gate of each auxiliary transistor ti is connected to the drain of each charge transfer transistor mi (i.e., node ni). The bulk of each auxiliary transistor ti is connected to the bulk of each charge transfer transistor mi, which is generally grounded. The auxiliary transistors ti and the PH1A and PH2A phases control the gate voltage of the charge transfer transistors mi.
The operation of the first stage of the pump is now explained, with all subsequent stages operating in the same manner. The operation commences with the PH1 phase starting to rise. Initially, charge transfer transistors m1 and m2 are non-conducting (i.e., turned off), since the PH1A and PH2A phases are in their low phase. The PH1 phase then fully rises and injects energy into node n1, raising (or “pushing”) node n1 to a voltage boosted above Vdd, such as 2Vdd. The rise of node n1 forces node g1 to Vdd through auxiliary transistor t1. Since the source of charge transfer transistor m1 is connected to Vdd at node n0, the gate-source voltage bias Vgs of charge transfer transistor m1 is zero, assuring that transistor m1 is turned off.
After a short time, typically in the order of several nanoseconds, the PH1A phase rises, which makes charge transfer transistor m2 conduct (i.e., turns on). During this time, node n1 is at a higher voltage than node n2. Since, as just mentioned, charge transfer transistor m2 is conducting, charge is transferred from node n1 to node n2. During the next phase, the PH2 phase rises and the PH1 phase drops. This causes node n1 to drop and node n2 to rise, thereby causing charge to be transferred from node n2 to node n3. In this manner charge is transferred along the pump. Each of the gi nodes is raised by a Vdd level with respect to the ni nodes when charge transfer is taking place. In the latter stages of the pump, the source and drain nodes (i.e., nodes n3 and n4) are raised well above the bulk, which is usually grounded.
The large voltage difference between the high source/drain voltages and the low bulk voltage causes a problem, called the body or bulk effect, which is now explained. (The terms body and bulk are used interchangeably throughout the specification and claims.)
Positive charge pumps generally use NMOS transistors, and this requires the body of the charge transfer transistors to be at the lowest voltage, in general ground (GND). (Negative charge pumps have the opposite requirement, and PMOS transistors are generally used.) However, in positive charge pumps there can be a significant loss of energy in the latter pump stages due to the “body effect”. In NMOS, the body effect is an increase in the threshold voltage (Vt), due to the fact that the bulk or body of the transistor is at a lower voltage than the source. Due to the body effect, the threshold voltage Vt of the NMOS transistors progressively increases from the stages near the input terminal of the charge pump to the stages near the output terminal. For example, in the prior art charge pump of FIG. 1, the threshold voltage Vt of charge transfer transistors mi progressively increases from transistor m1 to transistor m4. In transistor m4, as mentioned hereinabove, the source and drain nodes n3 and n4, have been raised well above the bulk. This reduces the efficiency of the charge pump, because, the voltage gain of each stage decreases, which means that a higher number of stages is necessary for generating a given voltage.
In some CMOS processes, such as triple-well and silicon-on-insulator (SOI), it is possible to raise the bulk of the NMOS charge transfer transistors above the grounded substrate, which would reduce the body effect by diminishing the voltage difference between the bulk and the source/drain. However, in the prior art, this entails certain risks. For example, if the bulk voltage is raised above the source or drain voltage, then parasitic bipolar transistors (typically used in CMOS circuitry) can turn on, which can cause either latchup or drain the charge from the pump.
In many circuits, not necessarily charge pumps, the bulk effect is eliminated by connecting the bulk node to the source node. This is not possible in a charge pump, however, because the “source” can be higher or lower than the “drain” by Vdd, depending upon the clock cycle. This would cause parasitic diodes to turn on, resulting in the unwanted bipolar transistor turn-on and latchup.
One method for compensating for the body effect is described in U.S. Pat. No. 6,064,251 to Park. Park uses charge pump stages coupled in series. Each charge pump stage has two clock terminals that receive two phase shifted clock signals. The charge pump stages are configured so that adjacent charge pump stages receive different clock signals. The phases of the clock signals are such that the pump elements are boosted well above the threshold voltage Vt, thereby providing the transistors with sufficient overdrive to transfer energy along the pump. However, clock boosting uses a significant amount of power consumption and is thus very wasteful.